refactor(scripts): use logging instead of print in the deploy credential guard #87

Context

Follow-up from #84. scripts/check_deploy_credentials.py (lines 155-157) writes its failure report with print(..., file=sys.stderr). AGENTS.md §2.6 says: "Standard logging via logging.getLogger(__name__). Never use print()."

The guard is a CLI script whose stderr output is intentionally user-facing, so the rule was not applied. It should still follow the repository standard.

Acceptance criteria

• Emit the header and one record per problem through a module logger configured with a stderr StreamHandler (no basicConfig side effects on import).
• Keep the exit code (1) and the existing message text so the compose failure stays readable.
• Update tests/unit/test_deploy_credentials.py::test_main_reports_problems_on_stderr to assert via caplog (or keep asserting captured stderr after configuring the handler).